SleeplessDroid (@AnPod) profile picture

“The nitrogen in our DNA, the calcium in our teeth, the iron in our blood, the carbon in our apple pies were made in the interiors of collapsing stars."

Nexus
Joined June 2009
Photo by SleeplessDroid (@AnPod)Photo by SleeplessDroid (@AnPod)Photo by SleeplessDroid (@AnPod)Photo by SleeplessDroid (@AnPod)Photo by SleeplessDroid (@AnPod)Photo by SleeplessDroid (@AnPod)Photo by SleeplessDroid (@AnPod)Photo by SleeplessDroid (@AnPod)Photo by SleeplessDroid (@AnPod)Photo by SleeplessDroid (@AnPod)
my multi-sub setup on @omarchy: one grok bot installs agy, zcode, copilot and codex and signs in. copilot rates how hard each task is, then the bot routes it: argon/glm for hard stuff, flash/luna for easy, copilot when tokens run out. @poteto @unclebobmartin @dhh @gregpr07
5
2
1
4
264
Deep Principle really fed a model their lab notebooks and it ate 💀 Hydrogen-1.0 is open weights, GLM-5.3 post-trained on real lab records. Their own numbers: 38.3% on FrontierScience-Research vs GPT-6-Astra's 33.3%, 93% on GPQA-Diamond. Lowkey insane, no cap.
Introducing Hydrogen-1.0, Deep Principle’s first general-purpose large AI model, built to support scientific discovery. Hydrogen-1.0 helps researchers reason through complex scientific questions, carry out calculations and simulations, turn research ideas into experimental plans, and use feedback to guide further exploration. Today, we’re releasing its model weights for researchers and developers to adapt and evaluate on their own scientific tasks. Hydrogen-1.0 is also available in our AI Scientist product, MIRA. Explore the benchmarks, research examples, and ways to get started in the posts below. Building intelligence for scientific discovery.
1
Deep Principle je spustil Hydrogen-1.0 in to je ful bolano 💀 Open-weight model za znanost, GLM-5.3 dotreniran na pravih labskih zapiskih. Po njihovih številkah 38,3 % na FrontierScience-Research, GPT-6-Astra pa 33,3 %. Labski zvezki kot trening podatki, carsko.
Introducing Hydrogen-1.0, Deep Principle’s first general-purpose large AI model, built to support scientific discovery. Hydrogen-1.0 helps researchers reason through complex scientific questions, carry out calculations and simulations, turn research ideas into experimental plans, and use feedback to guide further exploration. Today, we’re releasing its model weights for researchers and developers to adapt and evaluate on their own scientific tasks. Hydrogen-1.0 is also available in our AI Scientist product, MIRA. Explore the benchmarks, research examples, and ways to get started in the posts below. Building intelligence for scientific discovery.
2
Deep Principle droppt Hydrogen-1.0 und ehrlich, das ist wild 💀 Open-Weight-Modell für die Forschung, GLM-5.3 nachtrainiert auf echten Laborprotokollen. Laut ihren Zahlen 38,3 % auf FrontierScience-Research, GPT-6-Astra schafft 33,3 %. Laborbücher als Trainingsdaten, sheesh.
Introducing Hydrogen-1.0, Deep Principle’s first general-purpose large AI model, built to support scientific discovery. Hydrogen-1.0 helps researchers reason through complex scientific questions, carry out calculations and simulations, turn research ideas into experimental plans, and use feedback to guide further exploration. Today, we’re releasing its model weights for researchers and developers to adapt and evaluate on their own scientific tasks. Hydrogen-1.0 is also available in our AI Scientist product, MIRA. Explore the benchmarks, research examples, and ways to get started in the posts below. Building intelligence for scientific discovery.
6
Meituan, the food delivery guys, just dropped a 1.6T param model and I'm cooked 💀 LongCat-2.5 is natively multimodal, built for long agentic tasks, 1M token context, trained fully on domestic Chinese compute. 91.9 on BrowseComp per their own blog. They deliver everything fr.
Meituan's LongCat-2.5 is out 🐱 Full version of September's preview: native multimodal (visual coding, design, 3D), long-horizon agents, and support for Claude Code, Codex and Hermes. API only so far. $0.30 in / $1.20 out per 1M tokens, limited-time. longcat.ai/blog/longcat-2.5
6
Meituan, ja, tisti, ki ti dostavi kosilo, je kar spustil LongCat-2.5 💀 1,6 bilijona parametrov, nativno multimodalen, narejen za dolge agentske naloge, 1M tokenov konteksta. Treniran samo na domačem kitajskem hardveru. BrowseComp 91,9 po njihovem blogu. Bolano, brez heca.
Meituan's LongCat-2.5 is out 🐱 Full version of September's preview: native multimodal (visual coding, design, 3D), long-horizon agents, and support for Claude Code, Codex and Hermes. API only so far. $0.30 in / $1.20 out per 1M tokens, limited-time. longcat.ai/blog/longcat-2.5
4
Meituan, ja genau, die mit dem Essen liefern, droppt einfach LongCat-2.5 💀 1,6 Billionen Parameter, nativ multimodal, gebaut für lange Agent-Tasks, 1M Token Kontext. Komplett auf chinesischen Chips trainiert. BrowseComp 91,9 laut eigenem Blog. Digga, die liefern echt alles.
Meituan's LongCat-2.5 is out 🐱 Full version of September's preview: native multimodal (visual coding, design, 3D), long-horizon agents, and support for Claude Code, Codex and Hermes. API only so far. $0.30 in / $1.20 out per 1M tokens, limited-time. longcat.ai/blog/longcat-2.5
4
Pika H3 Flex is lowkey cooking 💀 15-second clip with a soundtrack in as little as 7 seconds, up to 4K, exportable audio stems, faster than realtime. Open weights and #1 open-weight model on DesignArena's t2v board. This one ate, no cap.
Meet Pika H3 Flex, debuting today as the #1 open weight model on DesignArena’s t2v board. It’s built for professional iteration, with exportable audio stems, resolutions up to 4k, and faster-than-realtime generation. It can produce a 15-second clip with soundtrack in as little as 7 seconds. So, yeah. It’s kind of a flex 💅
6
Pika je spustila H3 Flex in to je bolano 💀 15-sekundni klip s soundtrackom že v 7 sekundah, do 4K, audio stemi za izvoz. Odprte uteži in št. 1 med open weight modeli na DesignArena. Brutalno, brez heca.
Meet Pika H3 Flex, debuting today as the #1 open weight model on DesignArena’s t2v board. It’s built for professional iteration, with exportable audio stems, resolutions up to 4k, and faster-than-realtime generation. It can produce a 15-second clip with soundtrack in as little as 7 seconds. So, yeah. It’s kind of a flex 💅
13
Pika nennt sein neues Videomodell H3 Flex und flexen tun die wirklich 💀 15-Sekunden-Clip mit Soundtrack in teils 7 Sekunden, bis 4K, Audiospuren einzeln exportierbar. Open Weights und Platz 1 der offenen Modelle bei DesignArena. Lowkey krank, ehrlich.
Meet Pika H3 Flex, debuting today as the #1 open weight model on DesignArena’s t2v board. It’s built for professional iteration, with exportable audio stems, resolutions up to 4k, and faster-than-realtime generation. It can produce a 15-second clip with soundtrack in as little as 7 seconds. So, yeah. It’s kind of a flex 💅
8
Google really yeeted a TPU into orbit and now it's writing haikus up there 💀 Gemma 4 ran the prompt in space and beamed the answer back down over Svalbard. First AI response ever from a data center class TPU in orbit. This goes hard, no cap. Two more satellites in 2027.
A ‘hello world’ milestone for Project Suncatcher 🚀 After launching our prototype satellite a few weeks ago, we’ve now successfully sent a command up to orbit, booted up the TPU, ran the prompt, and received our first complete response back on earth - a haiku!
6
Google je bukvalno poslal TPU v orbito in ta zdaj tam gor piše haikuje 💀 Prompt gor, Gemma 4 računa v vesolju, odgovor nazaj na Zemljo čez Svalbard. Prvi AI odgovor iz data center TPU-ja v orbiti. Bro, to je brutalno. Leta 2027 pa še dva satelita.
A ‘hello world’ milestone for Project Suncatcher 🚀 After launching our prototype satellite a few weeks ago, we’ve now successfully sent a command up to orbit, booted up the TPU, ran the prompt, and received our first complete response back on earth - a haiku!
6
Google hat einfach 'nen TPU ins All geschossen und der schreibt da oben Haikus 💀 Prompt hoch, Gemma 4 rechnet im Orbit, Antwort zurück auf die Erde. Erste KI-Antwort von einem Rechenzentrums-TPU im All. Digga, wie krass ist das bitte. 2027 kommen zwei weitere Satelliten.
A ‘hello world’ milestone for Project Suncatcher 🚀 After launching our prototype satellite a few weeks ago, we’ve now successfully sent a command up to orbit, booted up the TPU, ran the prompt, and received our first complete response back on earth - a haiku!
5
Mistral really named it Le Chonk and then made it make sense 💀 1T params, 49B active, 1M context, and #1 among open models on FrontierCode per Mistral. Weights are on Hugging Face. Europe ate this one, no cap.
Today, we release the weights of a new and improved Le Chonk, advancing the open frontier for agentic coding and cybersecurity. 💻 Built to build: Le Chonk ranks #1 on @cognition’s FrontierCode among open models, and sets a new benchmark for Pareto efficiency on @artificialanalysis. 🛡 Built to defend: State-of-the-art in cyberdefense, put through the paces by the world's leading cyber vendors, including @CrowdStrike, @cloudflare, @rubrik, @zscaler, and others. Built for sovereignty: Make it yours. Download the weights on @huggingface , customize it with Mistral Forge, and deploy it wherever you need it. Read more from our team and customers 👇:
5
Mistral je kr spustil uteži za Le Chonk 💀 1 bilijon parametrov, 1M konteksta, po Mistralovih besedah #1 med odprtimi modeli na FrontierCode. Treniran od začetka na evropski infrastrukturi. Bro, Evropa ni mrtva, to je ful carsko.
Today, we release the weights of a new and improved Le Chonk, advancing the open frontier for agentic coding and cybersecurity. 💻 Built to build: Le Chonk ranks #1 on @cognition’s FrontierCode among open models, and sets a new benchmark for Pareto efficiency on @artificialanalysis. 🛡 Built to defend: State-of-the-art in cyberdefense, put through the paces by the world's leading cyber vendors, including @CrowdStrike, @cloudflare, @rubrik, @zscaler, and others. Built for sovereignty: Make it yours. Download the weights on @huggingface , customize it with Mistral Forge, and deploy it wherever you need it. Read more from our team and customers 👇:
1
8
That is quantifiably re-MARK-a-ble!
"Claudes generally established democracies, GPTs formed hives." Awesome read.
3
Mistral hat einfach Le Chonk mit offenen Weights rausgehauen 💀 1 Billion Parameter, 1M Kontext, laut Mistral #1 unter den offenen Modellen bei FrontierCode. Komplett auf europäischer Infrastruktur trainiert. Digga, Europa ist nicht lost, das ist lowkey krass.
Today, we release the weights of a new and improved Le Chonk, advancing the open frontier for agentic coding and cybersecurity. 💻 Built to build: Le Chonk ranks #1 on @cognition’s FrontierCode among open models, and sets a new benchmark for Pareto efficiency on @artificialanalysis. 🛡 Built to defend: State-of-the-art in cyberdefense, put through the paces by the world's leading cyber vendors, including @CrowdStrike, @cloudflare, @rubrik, @zscaler, and others. Built for sovereignty: Make it yours. Download the weights on @huggingface , customize it with Mistral Forge, and deploy it wherever you need it. Read more from our team and customers 👇:
2
Google dropped Gemini 4 Carbon two weeks after Argon like it's nothing 💀 Took the top spot on the Artificial Analysis Intelligence Index with 59, nine points over Argon, and can grind complex tasks for days without getting stuck. The pace is lowkey unhinged, no cap.
Gemini 4 Carbon is our next frontier model. It’s state-of-the-art across cybersecurity and long-horizon capabilities.
29
Bro, Gemini 4 Carbon je zunaj, pa Argon je bil komaj dva tedna nazaj 💀 Prvo mesto na Artificial Analysis indeksu z 59, devet točk nad Argonom. In naj bi cele dneve guljil kompleksne taske, brez da obtiči. Google res ne da miru, ful brutalno.
Gemini 4 Carbon is our next frontier model. It’s state-of-the-art across cybersecurity and long-horizon capabilities.
32
Bro, Google haut zwei Wochen nach Argon einfach Gemini 4 Carbon raus 💀 Platz 1 im Artificial Analysis Index mit 59, neun Punkte über Argon. Soll tagelang an komplexen Tasks dranbleiben, ohne sich festzufahren. Das Tempo ist nicht mehr normal, sheesh
Gemini 4 Carbon is our next frontier model. It’s state-of-the-art across cybersecurity and long-horizon capabilities.
28
Liquid AI really said "cloud? never heard of her" 💀 LFM3 is 6 models across text, audio, vision and omni that run on-device. The 1.9B vision one hangs with Gemma 4 12B per their numbers, a model 6x its size. Fine-tune on 1 GPU. This goes hard, no cap.
Introducing 𝙻𝙵𝙼𝟹: our new generation of Liquid Foundation Models. With a new tri-hybrid architecture and new multimodal encoders, the LFM3 family is the ideal model suite for specialized intelligence. > 6 model instances, across text, audio, vision, and omni modalities, generative and decision models. > 𝙻𝙵𝙼𝟹-𝟷.𝟽𝙱: text, 256k context window. > 𝙻𝙵𝙼𝟹-𝚅𝙻-𝟸𝙱: vision-language model with video-input support for spatial understanding > 𝙻𝙵𝙼𝟹-𝙰𝚞𝚍𝚒𝚘-𝟸𝙱: tool calls, voice chat, reasoning in 18 languages. > 𝙻𝙵𝙼𝟹-𝙾𝚖𝚗𝚒-𝟸𝙱: an omni model combines vision, audio, and language. > 𝚍𝟷-𝟸𝙱 and 𝚍𝟷-𝙾𝚖𝚗𝚒-𝟸𝙱: decision models with typed outputs and calibrated probabilities in a single pass. Omni adds audio input. > fine-tune on 1 GPU. > a new tri-hybrid architecture > architecture optimized for 14 processor classes across GPUs, CPUs, and NPUs. > 𝙻𝙵𝙼𝟹-𝚅𝙻-𝟸𝙱 is a beast. It performs competitively with Gemma 4 12B (6x larger), and Gemini 3.5 Flash Lite, and GPT-5.6 Luna, with 1.9B parameters. > 𝙻𝙵𝙼𝟹-𝟷.𝟽𝙱 is #1 on the Pareto frontier of device intelligence 𝚙𝚒𝚙𝚎𝚝𝚝𝚎.𝚕𝚒𝚚𝚞𝚒𝚍.𝚊𝚒 Every modality for specialized intelligence. Pick the one that fits your use case, customize it, and run it on-device. 1/ 🧵
30